Abstract: Aiming at the defects of long training time and slow speed due to redundant information and mass data in the support vector machine (SVM) based load forecasting and taking advantage of imprecise or incomplete knowledge and redundancy information handling by fuzzy rough sets (FRS), a short-term load forecasting model combining SVM and FRS is proposed, in which the attribute reduction algorithm of FRS is used to deal with the information bloat of the numerous power load affecting factors and eliminate the factors irrelevant to decision-making information; then, the simplified factors are input into SVM to conduct the forecasting. The simulation result shows that this forecasting model can ensure the prediction accuracy and speed up the calculation.

Key words: power system, short-term load forecasting, fuzzy rough set, attribute reduction algorithm, membership function, input variable selection, support vector machine, nonlinear regression